SlavikMIPT

[infra] Add cortex-m7 platform

2

This commit addds cortex-m7 to supported platforms. ONE-DCO-1.0-Signed-off-by: Vyacheslav Bazhenov [email protected]

This commit introduces luci-micro benchmark build for stm32f7.

This commit adds Dockerfile with environment for luci-micro build and test ONE-DCO-1.0-Signed-off-by: Vyacheslav Bazhenov

This commit adds support of UINT4 quantized weights in FullyConnected kernel. For #12804 ONE-DCO-1.0-Signed-off-by: Vyacheslav Bazhenov

[DRAFT][onert-micro] Introduce Huffman Transcoder

1

This commit introduces Huffman Transcoder for lossless weight compression. ONE-DCO-1.0-Signed-off-by: Vyacheslav Bazhenov

[DRAFT][circle-quantizer] Support int4 quantization

2

This commit supports int4 quantization. ONE-DCO-1.0-Signed-off-by: Vyacheslav Bazhenov

[luci] Introduce minimum MSE quantization

1

This commit introduces minimum MSE weights quantization algorithm. For https://github.com/Samsung/ONE/issues/12184 DRAFT #12240 ONE-DCO-1.0-Signed-off-by: Vyacheslav Bazhenov

[circle-quantizer] Minimum MSE Quantization

14

### What Let's support calculating of scale using [MSE minimization](https://arxiv.org/abs/1902.06822) (see part 3.4). ### Why We can improve accuracy of quantized models (especially low-bit quantized) by finding optimal approximation of...

### What Let's support sparse weights tensor compression.  ### Why We can reduce model binary size and speed up inference by operating only on the nonzero values in the...

[luci] Introduce MinimumMSE quantization algorithm

3

This commit introduces MinimumMSE quantization algorithm. For https://github.com/Samsung/ONE/issues/13677 ONE-DCO-1.0-Signed-off-by: Vyacheslav Bazhenov